An extension of Pedersen's viscosity model for saturated black oil systems

Abstract

This thesis presents a modification of Pedersen's corresponding states compositional viscosity model for black oil systems when no compositional data are available. This new model provides better estimates for oil viscosity than previously existing correlations. The model was developed from 324 sets of differential liberation data consisting of 2,232 observations. Data was gathered from a number of locations worldwide and covered a broad range of fluid compositions, pressures, and temperatures. Linear and nonparametric regression analyses were used to process and correlate the data. The new correlation, which retained most of the functional form of Pedersen's model, includes modifications that make the model more user-friendly and applicable to black oil simulator models. These modifications to Pedersen's model include use of n-decane as the reference fluid; considering the oil mixture as a single component with molecular weight and critical properties correlated to its density, which may be evaluated from correlations or experimental data; and addition of a functional dependence to the solution gas/oil ratio, formation volume factor, and gas specific gravity. Additionally, using 2,232 points, improved correlations are provided for the oil formation volume factor and the solution gas/oil ratio. This is a progression of the work of Verlade. One of the main advantages of our model is that it is linear in the parameters used for the viscosity prediction. It is therefore easier to recalibrate when needed for simple regression. We also investigated a functional relationship amongst pressure, volume, temperature (PVT) properties and a set of variables required to describe the viscosity correlation using a nonparametric regression analysis. The objective of this analysis was to determine which variables affect viscosity the most and determine the minimum number of variables needed to develop a suitable viscosity correlation. The minimum average absolute error for this analysis was 29±0.2% for the nonfitted nonparametric regression analysis obtained by using GRACE software with 42 coefficients and 13 variables. The average absolute errors obtained from our model using the linear regression analysis were much lower than those obtained from the nonparametric regression analysis obtained from GRACE.